孙杨, 舒清态, 黄金君, 等. 基于贝叶斯法估计龙竹人工林叶面积指数模型[J]. 西南林业大学学报(自然科学), 2022, 42(6): 114–121 . DOI: 10.11929/j.swfu.202109059
引用本文: 孙杨, 舒清态, 黄金君, 等. 基于贝叶斯法估计龙竹人工林叶面积指数模型[J]. 西南林业大学学报(自然科学), 2022, 42(6): 114–121 . DOI: 10.11929/j.swfu.202109059
Sun Yang, Shu Qingtai, Huang Jinjun, Xi Lei, Liu Yueling, Luo Hao. Estimation of Leaf Area Index Model of Dendrocalamus giganteus Plantation Based on Bayesian Method[J]. Journal of Southwest Forestry University, 2022, 42(6): 114-121. DOI: 10.11929/j.swfu.202109059
Citation: Sun Yang, Shu Qingtai, Huang Jinjun, Xi Lei, Liu Yueling, Luo Hao. Estimation of Leaf Area Index Model of Dendrocalamus giganteus Plantation Based on Bayesian Method[J]. Journal of Southwest Forestry University, 2022, 42(6): 114-121. DOI: 10.11929/j.swfu.202109059

基于贝叶斯法估计龙竹人工林叶面积指数模型

Estimation of Leaf Area Index Model of Dendrocalamus giganteus Plantation Based on Bayesian Method

  • 摘要: 以云南广泛栽培龙竹的沧源和新平2个县为研究区,以异速生长方程为LAI基础模型,结合73块样地数据,采用非线性最小二乘、有先验信息贝叶斯法和分层贝叶斯3种方法对模型参数进行拟合,运用决定系数(R2)、均方根误差(RMSE)和估测精度(E)指标对拟合效果进行评价。结果表明:当未引入随机效应变量时,传统的最小二乘法和有先验信息贝叶斯法的R2、RMSE和E分别为0.4875、0.0071、75.31%和0.4874、0.0070、75.31%;引入随机效应变量后,分层贝叶斯方法R2、RMSE和E分别为0.6733、0.0057、80.27%,估测效果较最小二乘方法和有先验信息的贝叶斯方法有较为明显的提高,R2提高了0.1858,RMSE降低了0.0014,E提高了4.96%。对于有明显地域差异的样本,分层贝叶斯方法能明显提升模型参数估测精度,适合中大尺度上采样数据的模型参数估测。

     

    Abstract: The research takes Cangyuan County and Xinping County where Dendrocalamus giganteus is widely cultivated in Yunnan as the study area. Taking the allometric growth equation as the LAI basic model, combined with the data of 73 sample plots on the ground, the model parameters were fitted by nonlinear least squares, Bayesian method with prior information and hierarchical Bayesian method. The fitting effect was evaluated by using the determination coefficient(R2), root mean square error(RMSE) and estimation accuracy(E). The results show that the R2, RMSE and E of the traditional least square method and Bayesian method with prior information are 0.4875, 0.0071, 75.31% and 0.4874, 0.0070, 75.31% respectively without adding random effect variables. After introducing the random effect variable, the R2, RMSE and E of the hierarchical Bayesian method are 0.6733, 0.0057 and 80.27%, respectively. The estimation effect is significantly improved compared with the least square method and the Bayesian method with prior information. R2 increases by 0.1858, RMSE decreases by 0.0014, and E increases by 4.96%. Therefore, for the samples with obvious regional differences, hierarchical Bayesian method can significantly improve the accuracy of model parameter estimation, which is suitable for the model parameter estimation of sampling data on medium and large scales.

     

/

返回文章
返回